OpenMetadata MCP Server

The OpenMetadata MCP that ships full CRUD across every entity type — including OM 1.12+ Data Contracts, Metrics, Search Index, API Collections, and API Endpoints that the embedded MCP doesn't cover yet.

172 tools, 4 workflow Prompts (lineage impact / DQ investigation / glossary bootstrap / owner reassign), 7 MCP Resources, and aggregations like lineage-impact (downstream blast-radius w/ owner notification list), quality-rollup (DQ status across a scope), and official Data Contract validation.

What it does that others don't

• OM 1.12+ entity coverage — Data Contracts, Metrics, Search Index, API Collections, API Endpoints, plus native contract validation/result retrieval. Not in the embedded MCP yet.
• Aggregation tools — lineage-impact answers "what breaks if I change/drop X?" by walking lineage + counting consumers + breaking down by entity type + resolving the owner union for change-mgmt notifications, in one call. get-domain-summary returns domain + 6 child entity types via /search/query with track_total_hits in one call instead of 7 sequential. get-table-summary folds table + lineage + sample-data + DQ similarly.
• Semantic search — semantic-search over OM 1.12+ vector index (POST /search/vector/query). Useful when keyword search misses synonyms.
• MCP Prompts (4) — lineage-impact-analysis, data-quality-investigation, glossary-term-bootstrap, owner-change-propagation. Workflow templates the model invokes directly.
• MCP Resources (7) — om://table/{fqn}, om://glossary-term/{fqn}, om://lineage/{type}/{fqn}, om://search/{query}, om://dashboard/{fqn}, om://pipeline/{fqn}, om://schema/{fqn}.
• Token-efficient by design — extractFields projection on 28 read tools (drops changeDescription/version/updatedBy/href noise — ~80% size reduction), OM_TOOLS/OM_DISABLE category toggles, search-tools meta-tool.
• Apps SDK card — lineage-impact renders as a blast-radius card on ChatGPT clients (downstream/upstream counts + type breakdown + top consumers + owners-to-notify) via _meta["openai/outputTemplate"]. Claude clients receive the same JSON content.
• stdio + Streamable HTTP — defaults to stdio. Set MCP_TRANSPORT=http for ChatGPT Apps SDK or remote clients (Bearer auth via MCP_HTTP_TOKEN).

Try this — 5 prompts

Connect the server to Claude Desktop or Claude Code, then paste any of these:

1. Lineage impact — "The payments.transactions table is being deprecated. List every dashboard, pipeline, and ML model that depends on it (upstream + downstream, depth 3)."
2. Data quality investigation — "Show all failing test cases from the last 7 days. Group by table, then by test type, with pass/fail counts."
3. Glossary bootstrap — "Create a payments glossary with these 8 terms: chargeback, refund, settlement, KYC, AML, transaction, customer-id, payment-method. Link related terms."
4. Owner reassign — "User taehee is leaving. List every entity (table/dashboard/pipeline/ML model) where they are owner. Then reassign all of them to team data-platform."
5. Domain summary — "Summarize the analytics domain: total tables/dashboards/pipelines/ML models, top 5 by recent updates, and the data products it owns."

When to use this vs OpenMetadata's embedded MCP

OpenMetadata 1.12+ ships an embedded MCP. They are complementary:

Use the embedded MCP for RBAC-aware governance with SSO. Use this server for bulk metadata operations, full entity CRUD parity, automation, and OM clusters older than 1.12.

Install

Claude Desktop

{
  "mcpServers": {
    "openmetadata": {
      "command": "npx",
      "args": ["-y", "@us-all/openmetadata-mcp"],
      "env": {
        "OPENMETADATA_HOST": "http://your-host:8585",
        "OPENMETADATA_TOKEN": "<jwt-bot-token>"
      }
    }
  }
}

Claude Code

claude mcp add openmetadata -s user \
  -e OPENMETADATA_HOST=http://your-host:8585 \
  -e OPENMETADATA_TOKEN=<jwt-bot-token> \
  -- npx -y @us-all/openmetadata-mcp

Docker

docker run --rm -i \
  -e OPENMETADATA_HOST=http://your-host:8585 \
  -e OPENMETADATA_TOKEN=<jwt-bot-token> \
  ghcr.io/us-all/openmetadata-mcp-server

Build from source

git clone https://github.com/us-all/openmetadata-mcp-server.git
cd openmetadata-mcp-server && pnpm install && pnpm build
node dist/index.js

Get a token

1. Open OpenMetadata UI → Settings → Bots
2. Create a new bot or use an existing one (ingestion-bot works)
3. Copy the JWT token

Configuration

Categories (10): search, core, discovery, governance, quality, services, admin, events, entities, meta (always-on).

When MCP_TRANSPORT=http: POST /mcp (Bearer-auth JSON-RPC) + GET /health (public liveness).

Token efficiency

extractFields adds another ~80–90% reduction on individual responses (e.g. get-table 8KB → 200B with extractFields: "name,columns.*.name,columns.*.dataType"). Auto-applied across 28 read tools.

// without
get-table { "id": "..." }

// with
get-table { "id": "...", "extractFields": "name,description,columns.*.name,columns.*.dataType" }

MCP Prompts (4)

Workflow templates available via MCP prompts/list:

• lineage-impact-analysis — given an entity, walk upstream + downstream lineage and rank by impact.
• data-quality-investigation — diff DQ test results across two windows; cluster failure modes.
• glossary-term-bootstrap — bulk-create a glossary with N related terms, link automatically.
• owner-change-propagation — find all entities owned by user X, propose batch reassignment.

MCP Resources

URI-based read-only access:

om://table/{fqn} (table + columns + owners + tags + joins), om://glossary-term/{fqn}, om://lineage/{type}/{fqn} (depth 3), om://search/{query} (top 10 keyword hits), om://dashboard/{fqn}, om://pipeline/{fqn} (with tasks), om://schema/{fqn}.

Tools (172)

10 categories. Use search-tools to discover at runtime; full list collapsed below.